An AI detector tries to answer "did a model write this?" A writing checker like SlopScore answers a different question: "which sentences make this sound like a model wrote it?" The second question is easier to answer well, and more useful if you're the one editing.
Most detectors (GPTZero, Originality.ai, Turnitin's AI indicator, Pangram and others) run the text through a trained classifier and return a probability. Some look at how predictable each word is; others are trained on large sets of human and AI text. They're improving, but they still share three problems:
That's why many universities tell staff not to use a detector score as the only evidence of misconduct.
SlopScore doesn't guess authorship. It looks for specific, well-documented habits of chatbot prose (the "it's not X, it's Y" contrast, "let's dive in", inflated words like "pivotal", rows of dramatic fragments, chatbot leftovers like "I hope this helps") and highlights each one with a fix. The rules are public, the same text always gets the same score, and it runs in your browser.
A person who writes "in today's fast-paced world" gets flagged too. That's on purpose: readers react to the phrase the same way no matter who wrote it.
| If you want to... | Use |
|---|---|
| Edit your own AI-assisted draft | A pattern checker |
| Review freelancer or agency content before it publishes | A pattern checker, plus a style guide |
| Keep docs and READMEs readable in CI | A pattern checker (SlopScore has a CLI and GitHub Action) |
| Screen a large volume of submissions for possible AI use | A detector, with a human reviewing anything flagged |
| Decide whether someone cheated | Neither on its own. Talk to the person. |
It means the text avoids the common tells. It doesn't prove a person wrote it. A well-edited AI draft can score 0, and that's fine. The goal is writing that reads well, not a verdict about who typed it.
Try the checker or browse all 22 patterns.